Re: [hybi] WebSockets : Question about masqued frames !

[Willy Tarreau <[e-mail]>]

On Sat, Jun 11, 2011 at 08:10:28AM +0200, Iñaki Baz Castillo wrote:
> 2011/6/9 Anthony Catel <[e-mail]>:
> > It's mainly because of proxy traversal (please read previous discussions
> > about cache poisoning & co)
> 
> IMHO this doesn't answer the original question:
> 
> > So why not simply imagine a mask whose evolutionary of the Salt was fixed at the start (why not from the handshake key) and whose encryption evolve based on the contents of the frames?
> 
> The author of the thread is not suggesting removing the client->server
> masking, but just make it predictable (for example from the WS
> handshake data) rather than having each frame its own masking key.
> Would be any (security) issue in the suggested case? I don't think so,
> but just wondering.

This was discussed in great detail in the past (as everything around masking).
The issue if the mask doesn't change is that the server knows it, so if it is
running malicious code (and so does the client), then it can tell the mask to
the client which will be able to emit predicted contents over the wire. So the
mask must change between two exchanges so that the client cannot guess it.

Willy